blob: 64b5f3a309f54cb9f6caa5a5b8aa79dfa21383e5 [file] [log] [blame]
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001/*
2 * This file is part of UBIFS.
3 *
4 * Copyright (C) 2006-2008 Nokia Corporation.
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 as published by
8 * the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * more details.
14 *
15 * You should have received a copy of the GNU General Public License along with
16 * this program; if not, write to the Free Software Foundation, Inc., 51
17 * Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18 *
19 * Authors: Artem Bityutskiy (Битюцкий Артём)
20 * Adrian Hunter
21 */
22
23/*
24 * This file implements UBIFS journal.
25 *
26 * The journal consists of 2 parts - the log and bud LEBs. The log has fixed
27 * length and position, while a bud logical eraseblock is any LEB in the main
28 * area. Buds contain file system data - data nodes, inode nodes, etc. The log
29 * contains only references to buds and some other stuff like commit
30 * start node. The idea is that when we commit the journal, we do
31 * not copy the data, the buds just become indexed. Since after the commit the
32 * nodes in bud eraseblocks become leaf nodes of the file system index tree, we
33 * use term "bud". Analogy is obvious, bud eraseblocks contain nodes which will
34 * become leafs in the future.
35 *
36 * The journal is multi-headed because we want to write data to the journal as
37 * optimally as possible. It is nice to have nodes belonging to the same inode
38 * in one LEB, so we may write data owned by different inodes to different
39 * journal heads, although at present only one data head is used.
40 *
41 * For recovery reasons, the base head contains all inode nodes, all directory
42 * entry nodes and all truncate nodes. This means that the other heads contain
43 * only data nodes.
44 *
45 * Bud LEBs may be half-indexed. For example, if the bud was not full at the
46 * time of commit, the bud is retained to continue to be used in the journal,
47 * even though the "front" of the LEB is now indexed. In that case, the log
48 * reference contains the offset where the bud starts for the purposes of the
49 * journal.
50 *
51 * The journal size has to be limited, because the larger is the journal, the
52 * longer it takes to mount UBIFS (scanning the journal) and the more memory it
53 * takes (indexing in the TNC).
54 *
55 * All the journal write operations like 'ubifs_jnl_update()' here, which write
56 * multiple UBIFS nodes to the journal at one go, are atomic with respect to
57 * unclean reboots. Should the unclean reboot happen, the recovery code drops
58 * all the nodes.
59 */
60
61#include "ubifs.h"
62
63/**
64 * zero_ino_node_unused - zero out unused fields of an on-flash inode node.
65 * @ino: the inode to zero out
66 */
67static inline void zero_ino_node_unused(struct ubifs_ino_node *ino)
68{
69 memset(ino->padding1, 0, 4);
70 memset(ino->padding2, 0, 26);
71}
72
73/**
74 * zero_dent_node_unused - zero out unused fields of an on-flash directory
75 * entry node.
76 * @dent: the directory entry to zero out
77 */
78static inline void zero_dent_node_unused(struct ubifs_dent_node *dent)
79{
80 dent->padding1 = 0;
81 memset(dent->padding2, 0, 4);
82}
83
84/**
85 * zero_data_node_unused - zero out unused fields of an on-flash data node.
86 * @data: the data node to zero out
87 */
88static inline void zero_data_node_unused(struct ubifs_data_node *data)
89{
90 memset(data->padding, 0, 2);
91}
92
93/**
94 * zero_trun_node_unused - zero out unused fields of an on-flash truncation
95 * node.
96 * @trun: the truncation node to zero out
97 */
98static inline void zero_trun_node_unused(struct ubifs_trun_node *trun)
99{
100 memset(trun->padding, 0, 12);
101}
102
103/**
104 * reserve_space - reserve space in the journal.
105 * @c: UBIFS file-system description object
106 * @jhead: journal head number
107 * @len: node length
108 *
109 * This function reserves space in journal head @head. If the reservation
110 * succeeded, the journal head stays locked and later has to be unlocked using
111 * 'release_head()'. 'write_node()' and 'write_head()' functions also unlock
112 * it. Returns zero in case of success, %-EAGAIN if commit has to be done, and
113 * other negative error codes in case of other failures.
114 */
115static int reserve_space(struct ubifs_info *c, int jhead, int len)
116{
Artem Bityutskiy3edaae72009-03-03 19:22:53 +0200117 int err = 0, err1, retries = 0, avail, lnum, offs, squeeze;
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300118 struct ubifs_wbuf *wbuf = &c->jheads[jhead].wbuf;
119
120 /*
121 * Typically, the base head has smaller nodes written to it, so it is
122 * better to try to allocate space at the ends of eraseblocks. This is
123 * what the squeeze parameter does.
124 */
125 squeeze = (jhead == BASEHD);
126again:
127 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
128
129 if (c->ro_media) {
130 err = -EROFS;
131 goto out_unlock;
132 }
133
134 avail = c->leb_size - wbuf->offs - wbuf->used;
135 if (wbuf->lnum != -1 && avail >= len)
136 return 0;
137
138 /*
139 * Write buffer wasn't seek'ed or there is no enough space - look for an
140 * LEB with some empty space.
141 */
Artem Bityutskiy3edaae72009-03-03 19:22:53 +0200142 lnum = ubifs_find_free_space(c, len, &offs, squeeze);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300143 if (lnum >= 0) {
144 /* Found an LEB, add it to the journal head */
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300145 err = ubifs_add_bud_to_log(c, jhead, lnum, offs);
146 if (err)
147 goto out_return;
148 /* A new bud was successfully allocated and added to the log */
149 goto out;
150 }
151
152 err = lnum;
153 if (err != -ENOSPC)
154 goto out_unlock;
155
156 /*
157 * No free space, we have to run garbage collector to make
158 * some. But the write-buffer mutex has to be unlocked because
159 * GC also takes it.
160 */
161 dbg_jnl("no free space jhead %d, run GC", jhead);
162 mutex_unlock(&wbuf->io_mutex);
163
164 lnum = ubifs_garbage_collect(c, 0);
165 if (lnum < 0) {
166 err = lnum;
167 if (err != -ENOSPC)
168 return err;
169
170 /*
171 * GC could not make a free LEB. But someone else may
172 * have allocated new bud for this journal head,
173 * because we dropped @wbuf->io_mutex, so try once
174 * again.
175 */
176 dbg_jnl("GC couldn't make a free LEB for jhead %d", jhead);
177 if (retries++ < 2) {
178 dbg_jnl("retry (%d)", retries);
179 goto again;
180 }
181
182 dbg_jnl("return -ENOSPC");
183 return err;
184 }
185
186 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
187 dbg_jnl("got LEB %d for jhead %d", lnum, jhead);
188 avail = c->leb_size - wbuf->offs - wbuf->used;
189
190 if (wbuf->lnum != -1 && avail >= len) {
191 /*
192 * Someone else has switched the journal head and we have
Frederik Schwarzer025dfda2008-10-16 19:02:37 +0200193 * enough space now. This happens when more than one process is
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300194 * trying to write to the same journal head at the same time.
195 */
196 dbg_jnl("return LEB %d back, already have LEB %d:%d",
197 lnum, wbuf->lnum, wbuf->offs + wbuf->used);
198 err = ubifs_return_leb(c, lnum);
199 if (err)
200 goto out_unlock;
201 return 0;
202 }
203
204 err = ubifs_add_bud_to_log(c, jhead, lnum, 0);
205 if (err)
206 goto out_return;
207 offs = 0;
208
209out:
Artem Bityutskiya50412e2009-01-06 19:54:02 +0200210 err = ubifs_wbuf_seek_nolock(wbuf, lnum, offs, wbuf->dtype);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300211 if (err)
212 goto out_unlock;
213
214 return 0;
215
216out_unlock:
217 mutex_unlock(&wbuf->io_mutex);
218 return err;
219
220out_return:
221 /* An error occurred and the LEB has to be returned to lprops */
222 ubifs_assert(err < 0);
223 err1 = ubifs_return_leb(c, lnum);
224 if (err1 && err == -EAGAIN)
225 /*
226 * Return original error code only if it is not %-EAGAIN,
227 * which is not really an error. Otherwise, return the error
228 * code of 'ubifs_return_leb()'.
229 */
230 err = err1;
231 mutex_unlock(&wbuf->io_mutex);
232 return err;
233}
234
235/**
236 * write_node - write node to a journal head.
237 * @c: UBIFS file-system description object
238 * @jhead: journal head
239 * @node: node to write
240 * @len: node length
241 * @lnum: LEB number written is returned here
242 * @offs: offset written is returned here
243 *
244 * This function writes a node to reserved space of journal head @jhead.
245 * Returns zero in case of success and a negative error code in case of
246 * failure.
247 */
248static int write_node(struct ubifs_info *c, int jhead, void *node, int len,
249 int *lnum, int *offs)
250{
251 struct ubifs_wbuf *wbuf = &c->jheads[jhead].wbuf;
252
253 ubifs_assert(jhead != GCHD);
254
255 *lnum = c->jheads[jhead].wbuf.lnum;
256 *offs = c->jheads[jhead].wbuf.offs + c->jheads[jhead].wbuf.used;
257
258 dbg_jnl("jhead %d, LEB %d:%d, len %d", jhead, *lnum, *offs, len);
259 ubifs_prepare_node(c, node, len, 0);
260
261 return ubifs_wbuf_write_nolock(wbuf, node, len);
262}
263
264/**
265 * write_head - write data to a journal head.
266 * @c: UBIFS file-system description object
267 * @jhead: journal head
268 * @buf: buffer to write
269 * @len: length to write
270 * @lnum: LEB number written is returned here
271 * @offs: offset written is returned here
272 * @sync: non-zero if the write-buffer has to by synchronized
273 *
274 * This function is the same as 'write_node()' but it does not assume the
275 * buffer it is writing is a node, so it does not prepare it (which means
276 * initializing common header and calculating CRC).
277 */
278static int write_head(struct ubifs_info *c, int jhead, void *buf, int len,
279 int *lnum, int *offs, int sync)
280{
281 int err;
282 struct ubifs_wbuf *wbuf = &c->jheads[jhead].wbuf;
283
284 ubifs_assert(jhead != GCHD);
285
286 *lnum = c->jheads[jhead].wbuf.lnum;
287 *offs = c->jheads[jhead].wbuf.offs + c->jheads[jhead].wbuf.used;
288 dbg_jnl("jhead %d, LEB %d:%d, len %d", jhead, *lnum, *offs, len);
289
290 err = ubifs_wbuf_write_nolock(wbuf, buf, len);
291 if (err)
292 return err;
293 if (sync)
294 err = ubifs_wbuf_sync_nolock(wbuf);
295 return err;
296}
297
298/**
299 * make_reservation - reserve journal space.
300 * @c: UBIFS file-system description object
301 * @jhead: journal head
302 * @len: how many bytes to reserve
303 *
304 * This function makes space reservation in journal head @jhead. The function
305 * takes the commit lock and locks the journal head, and the caller has to
306 * unlock the head and finish the reservation with 'finish_reservation()'.
307 * Returns zero in case of success and a negative error code in case of
308 * failure.
309 *
310 * Note, the journal head may be unlocked as soon as the data is written, while
311 * the commit lock has to be released after the data has been added to the
312 * TNC.
313 */
314static int make_reservation(struct ubifs_info *c, int jhead, int len)
315{
316 int err, cmt_retries = 0, nospc_retries = 0;
317
318again:
319 down_read(&c->commit_sem);
320 err = reserve_space(c, jhead, len);
321 if (!err)
322 return 0;
323 up_read(&c->commit_sem);
324
325 if (err == -ENOSPC) {
326 /*
327 * GC could not make any progress. We should try to commit
328 * once because it could make some dirty space and GC would
329 * make progress, so make the error -EAGAIN so that the below
330 * will commit and re-try.
331 */
332 if (nospc_retries++ < 2) {
333 dbg_jnl("no space, retry");
334 err = -EAGAIN;
335 }
336
337 /*
338 * This means that the budgeting is incorrect. We always have
339 * to be able to write to the media, because all operations are
340 * budgeted. Deletions are not budgeted, though, but we reserve
341 * an extra LEB for them.
342 */
343 }
344
345 if (err != -EAGAIN)
346 goto out;
347
348 /*
349 * -EAGAIN means that the journal is full or too large, or the above
350 * code wants to do one commit. Do this and re-try.
351 */
352 if (cmt_retries > 128) {
353 /*
354 * This should not happen unless the journal size limitations
355 * are too tough.
356 */
357 ubifs_err("stuck in space allocation");
358 err = -ENOSPC;
359 goto out;
360 } else if (cmt_retries > 32)
361 ubifs_warn("too many space allocation re-tries (%d)",
362 cmt_retries);
363
364 dbg_jnl("-EAGAIN, commit and retry (retried %d times)",
365 cmt_retries);
366 cmt_retries += 1;
367
368 err = ubifs_run_commit(c);
369 if (err)
370 return err;
371 goto again;
372
373out:
374 ubifs_err("cannot reserve %d bytes in jhead %d, error %d",
375 len, jhead, err);
376 if (err == -ENOSPC) {
377 /* This are some budgeting problems, print useful information */
378 down_write(&c->commit_sem);
379 spin_lock(&c->space_lock);
380 dbg_dump_stack();
381 dbg_dump_budg(c);
382 spin_unlock(&c->space_lock);
383 dbg_dump_lprops(c);
384 cmt_retries = dbg_check_lprops(c);
385 up_write(&c->commit_sem);
386 }
387 return err;
388}
389
390/**
391 * release_head - release a journal head.
392 * @c: UBIFS file-system description object
393 * @jhead: journal head
394 *
395 * This function releases journal head @jhead which was locked by
396 * the 'make_reservation()' function. It has to be called after each successful
397 * 'make_reservation()' invocation.
398 */
399static inline void release_head(struct ubifs_info *c, int jhead)
400{
401 mutex_unlock(&c->jheads[jhead].wbuf.io_mutex);
402}
403
404/**
405 * finish_reservation - finish a reservation.
406 * @c: UBIFS file-system description object
407 *
408 * This function finishes journal space reservation. It must be called after
409 * 'make_reservation()'.
410 */
411static void finish_reservation(struct ubifs_info *c)
412{
413 up_read(&c->commit_sem);
414}
415
416/**
417 * get_dent_type - translate VFS inode mode to UBIFS directory entry type.
418 * @mode: inode mode
419 */
420static int get_dent_type(int mode)
421{
422 switch (mode & S_IFMT) {
423 case S_IFREG:
424 return UBIFS_ITYPE_REG;
425 case S_IFDIR:
426 return UBIFS_ITYPE_DIR;
427 case S_IFLNK:
428 return UBIFS_ITYPE_LNK;
429 case S_IFBLK:
430 return UBIFS_ITYPE_BLK;
431 case S_IFCHR:
432 return UBIFS_ITYPE_CHR;
433 case S_IFIFO:
434 return UBIFS_ITYPE_FIFO;
435 case S_IFSOCK:
436 return UBIFS_ITYPE_SOCK;
437 default:
438 BUG();
439 }
440 return 0;
441}
442
443/**
444 * pack_inode - pack an inode node.
445 * @c: UBIFS file-system description object
446 * @ino: buffer in which to pack inode node
447 * @inode: inode to pack
448 * @last: indicates the last node of the group
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300449 */
450static void pack_inode(struct ubifs_info *c, struct ubifs_ino_node *ino,
Artem Bityutskiyfd6c6b52008-07-22 12:19:09 +0300451 const struct inode *inode, int last)
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300452{
Artem Bityutskiyfd6c6b52008-07-22 12:19:09 +0300453 int data_len = 0, last_reference = !inode->i_nlink;
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300454 struct ubifs_inode *ui = ubifs_inode(inode);
455
456 ino->ch.node_type = UBIFS_INO_NODE;
457 ino_key_init_flash(c, &ino->key, inode->i_ino);
458 ino->creat_sqnum = cpu_to_le64(ui->creat_sqnum);
459 ino->atime_sec = cpu_to_le64(inode->i_atime.tv_sec);
460 ino->atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
461 ino->ctime_sec = cpu_to_le64(inode->i_ctime.tv_sec);
462 ino->ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
463 ino->mtime_sec = cpu_to_le64(inode->i_mtime.tv_sec);
464 ino->mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
465 ino->uid = cpu_to_le32(inode->i_uid);
466 ino->gid = cpu_to_le32(inode->i_gid);
467 ino->mode = cpu_to_le32(inode->i_mode);
468 ino->flags = cpu_to_le32(ui->flags);
469 ino->size = cpu_to_le64(ui->ui_size);
470 ino->nlink = cpu_to_le32(inode->i_nlink);
471 ino->compr_type = cpu_to_le16(ui->compr_type);
472 ino->data_len = cpu_to_le32(ui->data_len);
473 ino->xattr_cnt = cpu_to_le32(ui->xattr_cnt);
474 ino->xattr_size = cpu_to_le32(ui->xattr_size);
475 ino->xattr_names = cpu_to_le32(ui->xattr_names);
476 zero_ino_node_unused(ino);
477
478 /*
479 * Drop the attached data if this is a deletion inode, the data is not
480 * needed anymore.
481 */
482 if (!last_reference) {
483 memcpy(ino->data, ui->data, ui->data_len);
484 data_len = ui->data_len;
485 }
486
487 ubifs_prep_grp_node(c, ino, UBIFS_INO_NODE_SZ + data_len, last);
488}
489
490/**
491 * mark_inode_clean - mark UBIFS inode as clean.
492 * @c: UBIFS file-system description object
493 * @ui: UBIFS inode to mark as clean
494 *
495 * This helper function marks UBIFS inode @ui as clean by cleaning the
496 * @ui->dirty flag and releasing its budget. Note, VFS may still treat the
497 * inode as dirty and try to write it back, but 'ubifs_write_inode()' would
498 * just do nothing.
499 */
500static void mark_inode_clean(struct ubifs_info *c, struct ubifs_inode *ui)
501{
502 if (ui->dirty)
503 ubifs_release_dirty_inode_budget(c, ui);
504 ui->dirty = 0;
505}
506
507/**
508 * ubifs_jnl_update - update inode.
509 * @c: UBIFS file-system description object
510 * @dir: parent inode or host inode in case of extended attributes
511 * @nm: directory entry name
512 * @inode: inode to update
513 * @deletion: indicates a directory entry deletion i.e unlink or rmdir
514 * @xent: non-zero if the directory entry is an extended attribute entry
515 *
516 * This function updates an inode by writing a directory entry (or extended
517 * attribute entry), the inode itself, and the parent directory inode (or the
518 * host inode) to the journal.
519 *
520 * The function writes the host inode @dir last, which is important in case of
521 * extended attributes. Indeed, then we guarantee that if the host inode gets
522 * synchronized (with 'fsync()'), and the write-buffer it sits in gets flushed,
523 * the extended attribute inode gets flushed too. And this is exactly what the
524 * user expects - synchronizing the host inode synchronizes its extended
525 * attributes. Similarly, this guarantees that if @dir is synchronized, its
526 * directory entry corresponding to @nm gets synchronized too.
527 *
528 * If the inode (@inode) or the parent directory (@dir) are synchronous, this
529 * function synchronizes the write-buffer.
530 *
531 * This function marks the @dir and @inode inodes as clean and returns zero on
532 * success. In case of failure, a negative error code is returned.
533 */
534int ubifs_jnl_update(struct ubifs_info *c, const struct inode *dir,
535 const struct qstr *nm, const struct inode *inode,
536 int deletion, int xent)
537{
538 int err, dlen, ilen, len, lnum, ino_offs, dent_offs;
539 int aligned_dlen, aligned_ilen, sync = IS_DIRSYNC(dir);
540 int last_reference = !!(deletion && inode->i_nlink == 0);
541 struct ubifs_inode *ui = ubifs_inode(inode);
542 struct ubifs_inode *dir_ui = ubifs_inode(dir);
543 struct ubifs_dent_node *dent;
544 struct ubifs_ino_node *ino;
545 union ubifs_key dent_key, ino_key;
546
547 dbg_jnl("ino %lu, dent '%.*s', data len %d in dir ino %lu",
548 inode->i_ino, nm->len, nm->name, ui->data_len, dir->i_ino);
549 ubifs_assert(dir_ui->data_len == 0);
550 ubifs_assert(mutex_is_locked(&dir_ui->ui_mutex));
551
552 dlen = UBIFS_DENT_NODE_SZ + nm->len + 1;
553 ilen = UBIFS_INO_NODE_SZ;
554
555 /*
556 * If the last reference to the inode is being deleted, then there is
557 * no need to attach and write inode data, it is being deleted anyway.
558 * And if the inode is being deleted, no need to synchronize
559 * write-buffer even if the inode is synchronous.
560 */
561 if (!last_reference) {
562 ilen += ui->data_len;
563 sync |= IS_SYNC(inode);
564 }
565
566 aligned_dlen = ALIGN(dlen, 8);
567 aligned_ilen = ALIGN(ilen, 8);
568 len = aligned_dlen + aligned_ilen + UBIFS_INO_NODE_SZ;
569 dent = kmalloc(len, GFP_NOFS);
570 if (!dent)
571 return -ENOMEM;
572
573 /* Make reservation before allocating sequence numbers */
574 err = make_reservation(c, BASEHD, len);
575 if (err)
576 goto out_free;
577
578 if (!xent) {
579 dent->ch.node_type = UBIFS_DENT_NODE;
580 dent_key_init(c, &dent_key, dir->i_ino, nm);
581 } else {
582 dent->ch.node_type = UBIFS_XENT_NODE;
583 xent_key_init(c, &dent_key, dir->i_ino, nm);
584 }
585
586 key_write(c, &dent_key, dent->key);
587 dent->inum = deletion ? 0 : cpu_to_le64(inode->i_ino);
588 dent->type = get_dent_type(inode->i_mode);
589 dent->nlen = cpu_to_le16(nm->len);
590 memcpy(dent->name, nm->name, nm->len);
591 dent->name[nm->len] = '\0';
592 zero_dent_node_unused(dent);
593 ubifs_prep_grp_node(c, dent, dlen, 0);
594
595 ino = (void *)dent + aligned_dlen;
Artem Bityutskiyfd6c6b52008-07-22 12:19:09 +0300596 pack_inode(c, ino, inode, 0);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300597 ino = (void *)ino + aligned_ilen;
Artem Bityutskiyfd6c6b52008-07-22 12:19:09 +0300598 pack_inode(c, ino, dir, 1);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300599
600 if (last_reference) {
601 err = ubifs_add_orphan(c, inode->i_ino);
602 if (err) {
603 release_head(c, BASEHD);
604 goto out_finish;
605 }
Artem Bityutskiyde94eb52008-07-22 13:06:20 +0300606 ui->del_cmtno = c->cmt_no;
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300607 }
608
609 err = write_head(c, BASEHD, dent, len, &lnum, &dent_offs, sync);
610 if (err)
611 goto out_release;
612 if (!sync) {
613 struct ubifs_wbuf *wbuf = &c->jheads[BASEHD].wbuf;
614
615 ubifs_wbuf_add_ino_nolock(wbuf, inode->i_ino);
616 ubifs_wbuf_add_ino_nolock(wbuf, dir->i_ino);
617 }
618 release_head(c, BASEHD);
619 kfree(dent);
620
621 if (deletion) {
622 err = ubifs_tnc_remove_nm(c, &dent_key, nm);
623 if (err)
624 goto out_ro;
625 err = ubifs_add_dirt(c, lnum, dlen);
626 } else
627 err = ubifs_tnc_add_nm(c, &dent_key, lnum, dent_offs, dlen, nm);
628 if (err)
629 goto out_ro;
630
631 /*
632 * Note, we do not remove the inode from TNC even if the last reference
633 * to it has just been deleted, because the inode may still be opened.
634 * Instead, the inode has been added to orphan lists and the orphan
635 * subsystem will take further care about it.
636 */
637 ino_key_init(c, &ino_key, inode->i_ino);
638 ino_offs = dent_offs + aligned_dlen;
639 err = ubifs_tnc_add(c, &ino_key, lnum, ino_offs, ilen);
640 if (err)
641 goto out_ro;
642
643 ino_key_init(c, &ino_key, dir->i_ino);
644 ino_offs += aligned_ilen;
645 err = ubifs_tnc_add(c, &ino_key, lnum, ino_offs, UBIFS_INO_NODE_SZ);
646 if (err)
647 goto out_ro;
648
649 finish_reservation(c);
650 spin_lock(&ui->ui_lock);
651 ui->synced_i_size = ui->ui_size;
652 spin_unlock(&ui->ui_lock);
653 mark_inode_clean(c, ui);
654 mark_inode_clean(c, dir_ui);
655 return 0;
656
657out_finish:
658 finish_reservation(c);
659out_free:
660 kfree(dent);
661 return err;
662
663out_release:
664 release_head(c, BASEHD);
665out_ro:
666 ubifs_ro_mode(c, err);
667 if (last_reference)
668 ubifs_delete_orphan(c, inode->i_ino);
669 finish_reservation(c);
670 return err;
671}
672
673/**
674 * ubifs_jnl_write_data - write a data node to the journal.
675 * @c: UBIFS file-system description object
676 * @inode: inode the data node belongs to
677 * @key: node key
678 * @buf: buffer to write
679 * @len: data length (must not exceed %UBIFS_BLOCK_SIZE)
680 *
681 * This function writes a data node to the journal. Returns %0 if the data node
682 * was successfully written, and a negative error code in case of failure.
683 */
684int ubifs_jnl_write_data(struct ubifs_info *c, const struct inode *inode,
685 const union ubifs_key *key, const void *buf, int len)
686{
687 struct ubifs_data_node *data;
688 int err, lnum, offs, compr_type, out_len;
689 int dlen = UBIFS_DATA_NODE_SZ + UBIFS_BLOCK_SIZE * WORST_COMPR_FACTOR;
690 struct ubifs_inode *ui = ubifs_inode(inode);
691
Artem Bityutskiye84461a2008-10-29 12:08:43 +0200692 dbg_jnl("ino %lu, blk %u, len %d, key %s",
693 (unsigned long)key_inum(c, key), key_block(c, key), len,
694 DBGKEY(key));
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300695 ubifs_assert(len <= UBIFS_BLOCK_SIZE);
696
697 data = kmalloc(dlen, GFP_NOFS);
698 if (!data)
699 return -ENOMEM;
700
701 data->ch.node_type = UBIFS_DATA_NODE;
702 key_write(c, key, &data->key);
703 data->size = cpu_to_le32(len);
704 zero_data_node_unused(data);
705
Artem Bityutskiya9f2fc02008-12-23 14:39:14 +0200706 if (!(ui->flags & UBIFS_COMPR_FL))
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300707 /* Compression is disabled for this inode */
708 compr_type = UBIFS_COMPR_NONE;
709 else
710 compr_type = ui->compr_type;
711
712 out_len = dlen - UBIFS_DATA_NODE_SZ;
713 ubifs_compress(buf, len, &data->data, &out_len, &compr_type);
714 ubifs_assert(out_len <= UBIFS_BLOCK_SIZE);
715
716 dlen = UBIFS_DATA_NODE_SZ + out_len;
717 data->compr_type = cpu_to_le16(compr_type);
718
719 /* Make reservation before allocating sequence numbers */
720 err = make_reservation(c, DATAHD, dlen);
721 if (err)
722 goto out_free;
723
724 err = write_node(c, DATAHD, data, dlen, &lnum, &offs);
725 if (err)
726 goto out_release;
727 ubifs_wbuf_add_ino_nolock(&c->jheads[DATAHD].wbuf, key_inum(c, key));
728 release_head(c, DATAHD);
729
730 err = ubifs_tnc_add(c, key, lnum, offs, dlen);
731 if (err)
732 goto out_ro;
733
734 finish_reservation(c);
735 kfree(data);
736 return 0;
737
738out_release:
739 release_head(c, DATAHD);
740out_ro:
741 ubifs_ro_mode(c, err);
742 finish_reservation(c);
743out_free:
744 kfree(data);
745 return err;
746}
747
748/**
749 * ubifs_jnl_write_inode - flush inode to the journal.
750 * @c: UBIFS file-system description object
751 * @inode: inode to flush
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300752 *
753 * This function writes inode @inode to the journal. If the inode is
754 * synchronous, it also synchronizes the write-buffer. Returns zero in case of
755 * success and a negative error code in case of failure.
756 */
Artem Bityutskiy1f286812008-07-22 12:06:13 +0300757int ubifs_jnl_write_inode(struct ubifs_info *c, const struct inode *inode)
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300758{
Artem Bityutskiy1f286812008-07-22 12:06:13 +0300759 int err, lnum, offs;
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300760 struct ubifs_ino_node *ino;
761 struct ubifs_inode *ui = ubifs_inode(inode);
Artem Bityutskiy1f286812008-07-22 12:06:13 +0300762 int sync = 0, len = UBIFS_INO_NODE_SZ, last_reference = !inode->i_nlink;
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300763
Artem Bityutskiy1f286812008-07-22 12:06:13 +0300764 dbg_jnl("ino %lu, nlink %u", inode->i_ino, inode->i_nlink);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300765
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300766 /*
767 * If the inode is being deleted, do not write the attached data. No
768 * need to synchronize the write-buffer either.
769 */
Artem Bityutskiy1f286812008-07-22 12:06:13 +0300770 if (!last_reference) {
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300771 len += ui->data_len;
772 sync = IS_SYNC(inode);
773 }
774 ino = kmalloc(len, GFP_NOFS);
775 if (!ino)
776 return -ENOMEM;
777
778 /* Make reservation before allocating sequence numbers */
779 err = make_reservation(c, BASEHD, len);
780 if (err)
781 goto out_free;
782
Artem Bityutskiyfd6c6b52008-07-22 12:19:09 +0300783 pack_inode(c, ino, inode, 1);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300784 err = write_head(c, BASEHD, ino, len, &lnum, &offs, sync);
785 if (err)
786 goto out_release;
787 if (!sync)
788 ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf,
789 inode->i_ino);
790 release_head(c, BASEHD);
791
Artem Bityutskiy1f286812008-07-22 12:06:13 +0300792 if (last_reference) {
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300793 err = ubifs_tnc_remove_ino(c, inode->i_ino);
794 if (err)
795 goto out_ro;
796 ubifs_delete_orphan(c, inode->i_ino);
797 err = ubifs_add_dirt(c, lnum, len);
798 } else {
799 union ubifs_key key;
800
801 ino_key_init(c, &key, inode->i_ino);
802 err = ubifs_tnc_add(c, &key, lnum, offs, len);
803 }
804 if (err)
805 goto out_ro;
806
807 finish_reservation(c);
808 spin_lock(&ui->ui_lock);
809 ui->synced_i_size = ui->ui_size;
810 spin_unlock(&ui->ui_lock);
811 kfree(ino);
812 return 0;
813
814out_release:
815 release_head(c, BASEHD);
816out_ro:
817 ubifs_ro_mode(c, err);
818 finish_reservation(c);
819out_free:
820 kfree(ino);
821 return err;
822}
823
824/**
Adrian Hunter7d62ff22008-07-23 15:48:39 +0300825 * ubifs_jnl_delete_inode - delete an inode.
Artem Bityutskiyde94eb52008-07-22 13:06:20 +0300826 * @c: UBIFS file-system description object
827 * @inode: inode to delete
828 *
829 * This function deletes inode @inode which includes removing it from orphans,
830 * deleting it from TNC and, in some cases, writing a deletion inode to the
831 * journal.
832 *
833 * When regular file inodes are unlinked or a directory inode is removed, the
Adrian Hunter7d62ff22008-07-23 15:48:39 +0300834 * 'ubifs_jnl_update()' function writes a corresponding deletion inode and
Artem Bityutskiyde94eb52008-07-22 13:06:20 +0300835 * direntry to the media, and adds the inode to orphans. After this, when the
836 * last reference to this inode has been dropped, this function is called. In
837 * general, it has to write one more deletion inode to the media, because if
838 * a commit happened between 'ubifs_jnl_update()' and
839 * 'ubifs_jnl_delete_inode()', the deletion inode is not in the journal
Adrian Hunter7d62ff22008-07-23 15:48:39 +0300840 * anymore, and in fact it might not be on the flash anymore, because it might
841 * have been garbage-collected already. And for optimization reasons UBIFS does
Artem Bityutskiyde94eb52008-07-22 13:06:20 +0300842 * not read the orphan area if it has been unmounted cleanly, so it would have
843 * no indication in the journal that there is a deleted inode which has to be
844 * removed from TNC.
845 *
846 * However, if there was no commit between 'ubifs_jnl_update()' and
847 * 'ubifs_jnl_delete_inode()', then there is no need to write the deletion
Adrian Hunter7d62ff22008-07-23 15:48:39 +0300848 * inode to the media for the second time. And this is quite a typical case.
Artem Bityutskiyde94eb52008-07-22 13:06:20 +0300849 *
850 * This function returns zero in case of success and a negative error code in
851 * case of failure.
852 */
853int ubifs_jnl_delete_inode(struct ubifs_info *c, const struct inode *inode)
854{
855 int err;
856 struct ubifs_inode *ui = ubifs_inode(inode);
857
858 ubifs_assert(inode->i_nlink == 0);
859
860 if (ui->del_cmtno != c->cmt_no)
861 /* A commit happened for sure */
862 return ubifs_jnl_write_inode(c, inode);
863
864 down_read(&c->commit_sem);
865 /*
866 * Check commit number again, because the first test has been done
867 * without @c->commit_sem, so a commit might have happened.
868 */
869 if (ui->del_cmtno != c->cmt_no) {
870 up_read(&c->commit_sem);
871 return ubifs_jnl_write_inode(c, inode);
872 }
873
Artem Bityutskiyde94eb52008-07-22 13:06:20 +0300874 err = ubifs_tnc_remove_ino(c, inode->i_ino);
875 if (err)
876 ubifs_ro_mode(c, err);
Adrian Hunterf7691082008-07-23 16:55:55 +0300877 else
878 ubifs_delete_orphan(c, inode->i_ino);
Artem Bityutskiyde94eb52008-07-22 13:06:20 +0300879 up_read(&c->commit_sem);
880 return err;
881}
882
883/**
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300884 * ubifs_jnl_rename - rename a directory entry.
885 * @c: UBIFS file-system description object
886 * @old_dir: parent inode of directory entry to rename
887 * @old_dentry: directory entry to rename
888 * @new_dir: parent inode of directory entry to rename
889 * @new_dentry: new directory entry (or directory entry to replace)
890 * @sync: non-zero if the write-buffer has to be synchronized
891 *
892 * This function implements the re-name operation which may involve writing up
893 * to 3 inodes and 2 directory entries. It marks the written inodes as clean
894 * and returns zero on success. In case of failure, a negative error code is
895 * returned.
896 */
897int ubifs_jnl_rename(struct ubifs_info *c, const struct inode *old_dir,
898 const struct dentry *old_dentry,
899 const struct inode *new_dir,
900 const struct dentry *new_dentry, int sync)
901{
902 void *p;
903 union ubifs_key key;
904 struct ubifs_dent_node *dent, *dent2;
905 int err, dlen1, dlen2, ilen, lnum, offs, len;
906 const struct inode *old_inode = old_dentry->d_inode;
907 const struct inode *new_inode = new_dentry->d_inode;
908 int aligned_dlen1, aligned_dlen2, plen = UBIFS_INO_NODE_SZ;
909 int last_reference = !!(new_inode && new_inode->i_nlink == 0);
910 int move = (old_dir != new_dir);
911 struct ubifs_inode *uninitialized_var(new_ui);
912
913 dbg_jnl("dent '%.*s' in dir ino %lu to dent '%.*s' in dir ino %lu",
914 old_dentry->d_name.len, old_dentry->d_name.name,
915 old_dir->i_ino, new_dentry->d_name.len,
916 new_dentry->d_name.name, new_dir->i_ino);
917 ubifs_assert(ubifs_inode(old_dir)->data_len == 0);
918 ubifs_assert(ubifs_inode(new_dir)->data_len == 0);
919 ubifs_assert(mutex_is_locked(&ubifs_inode(old_dir)->ui_mutex));
920 ubifs_assert(mutex_is_locked(&ubifs_inode(new_dir)->ui_mutex));
921
922 dlen1 = UBIFS_DENT_NODE_SZ + new_dentry->d_name.len + 1;
923 dlen2 = UBIFS_DENT_NODE_SZ + old_dentry->d_name.len + 1;
924 if (new_inode) {
925 new_ui = ubifs_inode(new_inode);
926 ubifs_assert(mutex_is_locked(&new_ui->ui_mutex));
927 ilen = UBIFS_INO_NODE_SZ;
928 if (!last_reference)
929 ilen += new_ui->data_len;
930 } else
931 ilen = 0;
932
933 aligned_dlen1 = ALIGN(dlen1, 8);
934 aligned_dlen2 = ALIGN(dlen2, 8);
935 len = aligned_dlen1 + aligned_dlen2 + ALIGN(ilen, 8) + ALIGN(plen, 8);
936 if (old_dir != new_dir)
937 len += plen;
938 dent = kmalloc(len, GFP_NOFS);
939 if (!dent)
940 return -ENOMEM;
941
942 /* Make reservation before allocating sequence numbers */
943 err = make_reservation(c, BASEHD, len);
944 if (err)
945 goto out_free;
946
947 /* Make new dent */
948 dent->ch.node_type = UBIFS_DENT_NODE;
949 dent_key_init_flash(c, &dent->key, new_dir->i_ino, &new_dentry->d_name);
950 dent->inum = cpu_to_le64(old_inode->i_ino);
951 dent->type = get_dent_type(old_inode->i_mode);
952 dent->nlen = cpu_to_le16(new_dentry->d_name.len);
953 memcpy(dent->name, new_dentry->d_name.name, new_dentry->d_name.len);
954 dent->name[new_dentry->d_name.len] = '\0';
955 zero_dent_node_unused(dent);
956 ubifs_prep_grp_node(c, dent, dlen1, 0);
957
958 /* Make deletion dent */
959 dent2 = (void *)dent + aligned_dlen1;
960 dent2->ch.node_type = UBIFS_DENT_NODE;
961 dent_key_init_flash(c, &dent2->key, old_dir->i_ino,
962 &old_dentry->d_name);
963 dent2->inum = 0;
964 dent2->type = DT_UNKNOWN;
965 dent2->nlen = cpu_to_le16(old_dentry->d_name.len);
966 memcpy(dent2->name, old_dentry->d_name.name, old_dentry->d_name.len);
967 dent2->name[old_dentry->d_name.len] = '\0';
968 zero_dent_node_unused(dent2);
969 ubifs_prep_grp_node(c, dent2, dlen2, 0);
970
971 p = (void *)dent2 + aligned_dlen2;
972 if (new_inode) {
Artem Bityutskiyfd6c6b52008-07-22 12:19:09 +0300973 pack_inode(c, p, new_inode, 0);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300974 p += ALIGN(ilen, 8);
975 }
976
977 if (!move)
Artem Bityutskiyfd6c6b52008-07-22 12:19:09 +0300978 pack_inode(c, p, old_dir, 1);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300979 else {
Artem Bityutskiyfd6c6b52008-07-22 12:19:09 +0300980 pack_inode(c, p, old_dir, 0);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300981 p += ALIGN(plen, 8);
Artem Bityutskiyfd6c6b52008-07-22 12:19:09 +0300982 pack_inode(c, p, new_dir, 1);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300983 }
984
985 if (last_reference) {
986 err = ubifs_add_orphan(c, new_inode->i_ino);
987 if (err) {
988 release_head(c, BASEHD);
989 goto out_finish;
990 }
Artem Bityutskiyde94eb52008-07-22 13:06:20 +0300991 new_ui->del_cmtno = c->cmt_no;
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300992 }
993
994 err = write_head(c, BASEHD, dent, len, &lnum, &offs, sync);
995 if (err)
996 goto out_release;
997 if (!sync) {
998 struct ubifs_wbuf *wbuf = &c->jheads[BASEHD].wbuf;
999
1000 ubifs_wbuf_add_ino_nolock(wbuf, new_dir->i_ino);
1001 ubifs_wbuf_add_ino_nolock(wbuf, old_dir->i_ino);
1002 if (new_inode)
1003 ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf,
1004 new_inode->i_ino);
1005 }
1006 release_head(c, BASEHD);
1007
1008 dent_key_init(c, &key, new_dir->i_ino, &new_dentry->d_name);
1009 err = ubifs_tnc_add_nm(c, &key, lnum, offs, dlen1, &new_dentry->d_name);
1010 if (err)
1011 goto out_ro;
1012
1013 err = ubifs_add_dirt(c, lnum, dlen2);
1014 if (err)
1015 goto out_ro;
1016
1017 dent_key_init(c, &key, old_dir->i_ino, &old_dentry->d_name);
1018 err = ubifs_tnc_remove_nm(c, &key, &old_dentry->d_name);
1019 if (err)
1020 goto out_ro;
1021
1022 offs += aligned_dlen1 + aligned_dlen2;
1023 if (new_inode) {
1024 ino_key_init(c, &key, new_inode->i_ino);
1025 err = ubifs_tnc_add(c, &key, lnum, offs, ilen);
1026 if (err)
1027 goto out_ro;
1028 offs += ALIGN(ilen, 8);
1029 }
1030
1031 ino_key_init(c, &key, old_dir->i_ino);
1032 err = ubifs_tnc_add(c, &key, lnum, offs, plen);
1033 if (err)
1034 goto out_ro;
1035
1036 if (old_dir != new_dir) {
1037 offs += ALIGN(plen, 8);
1038 ino_key_init(c, &key, new_dir->i_ino);
1039 err = ubifs_tnc_add(c, &key, lnum, offs, plen);
1040 if (err)
1041 goto out_ro;
1042 }
1043
1044 finish_reservation(c);
1045 if (new_inode) {
1046 mark_inode_clean(c, new_ui);
1047 spin_lock(&new_ui->ui_lock);
1048 new_ui->synced_i_size = new_ui->ui_size;
1049 spin_unlock(&new_ui->ui_lock);
1050 }
1051 mark_inode_clean(c, ubifs_inode(old_dir));
1052 if (move)
1053 mark_inode_clean(c, ubifs_inode(new_dir));
1054 kfree(dent);
1055 return 0;
1056
1057out_release:
1058 release_head(c, BASEHD);
1059out_ro:
1060 ubifs_ro_mode(c, err);
1061 if (last_reference)
1062 ubifs_delete_orphan(c, new_inode->i_ino);
1063out_finish:
1064 finish_reservation(c);
1065out_free:
1066 kfree(dent);
1067 return err;
1068}
1069
1070/**
1071 * recomp_data_node - re-compress a truncated data node.
1072 * @dn: data node to re-compress
1073 * @new_len: new length
1074 *
1075 * This function is used when an inode is truncated and the last data node of
1076 * the inode has to be re-compressed and re-written.
1077 */
1078static int recomp_data_node(struct ubifs_data_node *dn, int *new_len)
1079{
1080 void *buf;
1081 int err, len, compr_type, out_len;
1082
1083 out_len = le32_to_cpu(dn->size);
1084 buf = kmalloc(out_len * WORST_COMPR_FACTOR, GFP_NOFS);
1085 if (!buf)
1086 return -ENOMEM;
1087
1088 len = le32_to_cpu(dn->ch.len) - UBIFS_DATA_NODE_SZ;
1089 compr_type = le16_to_cpu(dn->compr_type);
1090 err = ubifs_decompress(&dn->data, len, buf, &out_len, compr_type);
1091 if (err)
1092 goto out;
1093
1094 ubifs_compress(buf, *new_len, &dn->data, &out_len, &compr_type);
1095 ubifs_assert(out_len <= UBIFS_BLOCK_SIZE);
1096 dn->compr_type = cpu_to_le16(compr_type);
1097 dn->size = cpu_to_le32(*new_len);
1098 *new_len = UBIFS_DATA_NODE_SZ + out_len;
1099out:
1100 kfree(buf);
1101 return err;
1102}
1103
1104/**
1105 * ubifs_jnl_truncate - update the journal for a truncation.
1106 * @c: UBIFS file-system description object
1107 * @inode: inode to truncate
1108 * @old_size: old size
1109 * @new_size: new size
1110 *
1111 * When the size of a file decreases due to truncation, a truncation node is
1112 * written, the journal tree is updated, and the last data block is re-written
1113 * if it has been affected. The inode is also updated in order to synchronize
1114 * the new inode size.
1115 *
1116 * This function marks the inode as clean and returns zero on success. In case
1117 * of failure, a negative error code is returned.
1118 */
1119int ubifs_jnl_truncate(struct ubifs_info *c, const struct inode *inode,
1120 loff_t old_size, loff_t new_size)
1121{
1122 union ubifs_key key, to_key;
1123 struct ubifs_ino_node *ino;
1124 struct ubifs_trun_node *trun;
1125 struct ubifs_data_node *uninitialized_var(dn);
1126 int err, dlen, len, lnum, offs, bit, sz, sync = IS_SYNC(inode);
1127 struct ubifs_inode *ui = ubifs_inode(inode);
1128 ino_t inum = inode->i_ino;
1129 unsigned int blk;
1130
Artem Bityutskiye84461a2008-10-29 12:08:43 +02001131 dbg_jnl("ino %lu, size %lld -> %lld",
1132 (unsigned long)inum, old_size, new_size);
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001133 ubifs_assert(!ui->data_len);
1134 ubifs_assert(S_ISREG(inode->i_mode));
1135 ubifs_assert(mutex_is_locked(&ui->ui_mutex));
1136
1137 sz = UBIFS_TRUN_NODE_SZ + UBIFS_INO_NODE_SZ +
1138 UBIFS_MAX_DATA_NODE_SZ * WORST_COMPR_FACTOR;
1139 ino = kmalloc(sz, GFP_NOFS);
1140 if (!ino)
1141 return -ENOMEM;
1142
1143 trun = (void *)ino + UBIFS_INO_NODE_SZ;
1144 trun->ch.node_type = UBIFS_TRUN_NODE;
1145 trun->inum = cpu_to_le32(inum);
1146 trun->old_size = cpu_to_le64(old_size);
1147 trun->new_size = cpu_to_le64(new_size);
1148 zero_trun_node_unused(trun);
1149
1150 dlen = new_size & (UBIFS_BLOCK_SIZE - 1);
1151 if (dlen) {
1152 /* Get last data block so it can be truncated */
1153 dn = (void *)trun + UBIFS_TRUN_NODE_SZ;
1154 blk = new_size >> UBIFS_BLOCK_SHIFT;
1155 data_key_init(c, &key, inum, blk);
1156 dbg_jnl("last block key %s", DBGKEY(&key));
1157 err = ubifs_tnc_lookup(c, &key, dn);
1158 if (err == -ENOENT)
1159 dlen = 0; /* Not found (so it is a hole) */
1160 else if (err)
1161 goto out_free;
1162 else {
1163 if (le32_to_cpu(dn->size) <= dlen)
1164 dlen = 0; /* Nothing to do */
1165 else {
1166 int compr_type = le16_to_cpu(dn->compr_type);
1167
1168 if (compr_type != UBIFS_COMPR_NONE) {
1169 err = recomp_data_node(dn, &dlen);
1170 if (err)
1171 goto out_free;
1172 } else {
1173 dn->size = cpu_to_le32(dlen);
1174 dlen += UBIFS_DATA_NODE_SZ;
1175 }
1176 zero_data_node_unused(dn);
1177 }
1178 }
1179 }
1180
1181 /* Must make reservation before allocating sequence numbers */
1182 len = UBIFS_TRUN_NODE_SZ + UBIFS_INO_NODE_SZ;
1183 if (dlen)
1184 len += dlen;
1185 err = make_reservation(c, BASEHD, len);
1186 if (err)
1187 goto out_free;
1188
Artem Bityutskiyfd6c6b52008-07-22 12:19:09 +03001189 pack_inode(c, ino, inode, 0);
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001190 ubifs_prep_grp_node(c, trun, UBIFS_TRUN_NODE_SZ, dlen ? 0 : 1);
1191 if (dlen)
1192 ubifs_prep_grp_node(c, dn, dlen, 1);
1193
1194 err = write_head(c, BASEHD, ino, len, &lnum, &offs, sync);
1195 if (err)
1196 goto out_release;
1197 if (!sync)
1198 ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf, inum);
1199 release_head(c, BASEHD);
1200
1201 if (dlen) {
1202 sz = offs + UBIFS_INO_NODE_SZ + UBIFS_TRUN_NODE_SZ;
1203 err = ubifs_tnc_add(c, &key, lnum, sz, dlen);
1204 if (err)
1205 goto out_ro;
1206 }
1207
1208 ino_key_init(c, &key, inum);
1209 err = ubifs_tnc_add(c, &key, lnum, offs, UBIFS_INO_NODE_SZ);
1210 if (err)
1211 goto out_ro;
1212
1213 err = ubifs_add_dirt(c, lnum, UBIFS_TRUN_NODE_SZ);
1214 if (err)
1215 goto out_ro;
1216
1217 bit = new_size & (UBIFS_BLOCK_SIZE - 1);
1218 blk = (new_size >> UBIFS_BLOCK_SHIFT) + (bit ? 1 : 0);
1219 data_key_init(c, &key, inum, blk);
1220
1221 bit = old_size & (UBIFS_BLOCK_SIZE - 1);
Artem Bityutskiyf92b9822008-12-28 11:34:26 +02001222 blk = (old_size >> UBIFS_BLOCK_SHIFT) - (bit ? 0 : 1);
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001223 data_key_init(c, &to_key, inum, blk);
1224
1225 err = ubifs_tnc_remove_range(c, &key, &to_key);
1226 if (err)
1227 goto out_ro;
1228
1229 finish_reservation(c);
1230 spin_lock(&ui->ui_lock);
1231 ui->synced_i_size = ui->ui_size;
1232 spin_unlock(&ui->ui_lock);
1233 mark_inode_clean(c, ui);
1234 kfree(ino);
1235 return 0;
1236
1237out_release:
1238 release_head(c, BASEHD);
1239out_ro:
1240 ubifs_ro_mode(c, err);
1241 finish_reservation(c);
1242out_free:
1243 kfree(ino);
1244 return err;
1245}
1246
1247#ifdef CONFIG_UBIFS_FS_XATTR
1248
1249/**
1250 * ubifs_jnl_delete_xattr - delete an extended attribute.
1251 * @c: UBIFS file-system description object
1252 * @host: host inode
1253 * @inode: extended attribute inode
1254 * @nm: extended attribute entry name
1255 *
1256 * This function delete an extended attribute which is very similar to
1257 * un-linking regular files - it writes a deletion xentry, a deletion inode and
1258 * updates the target inode. Returns zero in case of success and a negative
1259 * error code in case of failure.
1260 */
1261int ubifs_jnl_delete_xattr(struct ubifs_info *c, const struct inode *host,
1262 const struct inode *inode, const struct qstr *nm)
1263{
1264 int err, xlen, hlen, len, lnum, xent_offs, aligned_xlen;
1265 struct ubifs_dent_node *xent;
1266 struct ubifs_ino_node *ino;
1267 union ubifs_key xent_key, key1, key2;
1268 int sync = IS_DIRSYNC(host);
1269 struct ubifs_inode *host_ui = ubifs_inode(host);
1270
1271 dbg_jnl("host %lu, xattr ino %lu, name '%s', data len %d",
1272 host->i_ino, inode->i_ino, nm->name,
1273 ubifs_inode(inode)->data_len);
1274 ubifs_assert(inode->i_nlink == 0);
1275 ubifs_assert(mutex_is_locked(&host_ui->ui_mutex));
1276
1277 /*
1278 * Since we are deleting the inode, we do not bother to attach any data
1279 * to it and assume its length is %UBIFS_INO_NODE_SZ.
1280 */
1281 xlen = UBIFS_DENT_NODE_SZ + nm->len + 1;
1282 aligned_xlen = ALIGN(xlen, 8);
1283 hlen = host_ui->data_len + UBIFS_INO_NODE_SZ;
1284 len = aligned_xlen + UBIFS_INO_NODE_SZ + ALIGN(hlen, 8);
1285
1286 xent = kmalloc(len, GFP_NOFS);
1287 if (!xent)
1288 return -ENOMEM;
1289
1290 /* Make reservation before allocating sequence numbers */
1291 err = make_reservation(c, BASEHD, len);
1292 if (err) {
1293 kfree(xent);
1294 return err;
1295 }
1296
1297 xent->ch.node_type = UBIFS_XENT_NODE;
1298 xent_key_init(c, &xent_key, host->i_ino, nm);
1299 key_write(c, &xent_key, xent->key);
1300 xent->inum = 0;
1301 xent->type = get_dent_type(inode->i_mode);
1302 xent->nlen = cpu_to_le16(nm->len);
1303 memcpy(xent->name, nm->name, nm->len);
1304 xent->name[nm->len] = '\0';
1305 zero_dent_node_unused(xent);
1306 ubifs_prep_grp_node(c, xent, xlen, 0);
1307
1308 ino = (void *)xent + aligned_xlen;
Artem Bityutskiyfd6c6b52008-07-22 12:19:09 +03001309 pack_inode(c, ino, inode, 0);
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001310 ino = (void *)ino + UBIFS_INO_NODE_SZ;
Artem Bityutskiyfd6c6b52008-07-22 12:19:09 +03001311 pack_inode(c, ino, host, 1);
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001312
1313 err = write_head(c, BASEHD, xent, len, &lnum, &xent_offs, sync);
1314 if (!sync && !err)
1315 ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf, host->i_ino);
1316 release_head(c, BASEHD);
1317 kfree(xent);
1318 if (err)
1319 goto out_ro;
1320
1321 /* Remove the extended attribute entry from TNC */
1322 err = ubifs_tnc_remove_nm(c, &xent_key, nm);
1323 if (err)
1324 goto out_ro;
1325 err = ubifs_add_dirt(c, lnum, xlen);
1326 if (err)
1327 goto out_ro;
1328
1329 /*
1330 * Remove all nodes belonging to the extended attribute inode from TNC.
1331 * Well, there actually must be only one node - the inode itself.
1332 */
1333 lowest_ino_key(c, &key1, inode->i_ino);
1334 highest_ino_key(c, &key2, inode->i_ino);
1335 err = ubifs_tnc_remove_range(c, &key1, &key2);
1336 if (err)
1337 goto out_ro;
1338 err = ubifs_add_dirt(c, lnum, UBIFS_INO_NODE_SZ);
1339 if (err)
1340 goto out_ro;
1341
1342 /* And update TNC with the new host inode position */
1343 ino_key_init(c, &key1, host->i_ino);
1344 err = ubifs_tnc_add(c, &key1, lnum, xent_offs + len - hlen, hlen);
1345 if (err)
1346 goto out_ro;
1347
1348 finish_reservation(c);
1349 spin_lock(&host_ui->ui_lock);
1350 host_ui->synced_i_size = host_ui->ui_size;
1351 spin_unlock(&host_ui->ui_lock);
1352 mark_inode_clean(c, host_ui);
1353 return 0;
1354
1355out_ro:
1356 ubifs_ro_mode(c, err);
1357 finish_reservation(c);
1358 return err;
1359}
1360
1361/**
1362 * ubifs_jnl_change_xattr - change an extended attribute.
1363 * @c: UBIFS file-system description object
1364 * @inode: extended attribute inode
1365 * @host: host inode
1366 *
1367 * This function writes the updated version of an extended attribute inode and
Artem Bityutskiy7d4e9cc2009-03-20 19:11:12 +02001368 * the host inode to the journal (to the base head). The host inode is written
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001369 * after the extended attribute inode in order to guarantee that the extended
1370 * attribute will be flushed when the inode is synchronized by 'fsync()' and
1371 * consequently, the write-buffer is synchronized. This function returns zero
1372 * in case of success and a negative error code in case of failure.
1373 */
1374int ubifs_jnl_change_xattr(struct ubifs_info *c, const struct inode *inode,
1375 const struct inode *host)
1376{
1377 int err, len1, len2, aligned_len, aligned_len1, lnum, offs;
Artem Bityutskiyc78c7e32008-08-12 16:30:12 +03001378 struct ubifs_inode *host_ui = ubifs_inode(host);
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001379 struct ubifs_ino_node *ino;
1380 union ubifs_key key;
1381 int sync = IS_DIRSYNC(host);
1382
1383 dbg_jnl("ino %lu, ino %lu", host->i_ino, inode->i_ino);
1384 ubifs_assert(host->i_nlink > 0);
1385 ubifs_assert(inode->i_nlink > 0);
1386 ubifs_assert(mutex_is_locked(&host_ui->ui_mutex));
1387
1388 len1 = UBIFS_INO_NODE_SZ + host_ui->data_len;
1389 len2 = UBIFS_INO_NODE_SZ + ubifs_inode(inode)->data_len;
1390 aligned_len1 = ALIGN(len1, 8);
1391 aligned_len = aligned_len1 + ALIGN(len2, 8);
1392
1393 ino = kmalloc(aligned_len, GFP_NOFS);
1394 if (!ino)
1395 return -ENOMEM;
1396
1397 /* Make reservation before allocating sequence numbers */
1398 err = make_reservation(c, BASEHD, aligned_len);
1399 if (err)
1400 goto out_free;
1401
Artem Bityutskiyfd6c6b52008-07-22 12:19:09 +03001402 pack_inode(c, ino, host, 0);
1403 pack_inode(c, (void *)ino + aligned_len1, inode, 1);
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001404
1405 err = write_head(c, BASEHD, ino, aligned_len, &lnum, &offs, 0);
1406 if (!sync && !err) {
1407 struct ubifs_wbuf *wbuf = &c->jheads[BASEHD].wbuf;
1408
1409 ubifs_wbuf_add_ino_nolock(wbuf, host->i_ino);
1410 ubifs_wbuf_add_ino_nolock(wbuf, inode->i_ino);
1411 }
1412 release_head(c, BASEHD);
1413 if (err)
1414 goto out_ro;
1415
1416 ino_key_init(c, &key, host->i_ino);
1417 err = ubifs_tnc_add(c, &key, lnum, offs, len1);
1418 if (err)
1419 goto out_ro;
1420
1421 ino_key_init(c, &key, inode->i_ino);
1422 err = ubifs_tnc_add(c, &key, lnum, offs + aligned_len1, len2);
1423 if (err)
1424 goto out_ro;
1425
1426 finish_reservation(c);
1427 spin_lock(&host_ui->ui_lock);
1428 host_ui->synced_i_size = host_ui->ui_size;
1429 spin_unlock(&host_ui->ui_lock);
1430 mark_inode_clean(c, host_ui);
1431 kfree(ino);
1432 return 0;
1433
1434out_ro:
1435 ubifs_ro_mode(c, err);
1436 finish_reservation(c);
1437out_free:
1438 kfree(ino);
1439 return err;
1440}
1441
1442#endif /* CONFIG_UBIFS_FS_XATTR */